An Adversarial Search Method Based on an Iterative Optimal Strategy

A deeper game-tree search can yield a higher decision quality in a heuristic minimax algorithm. However, exceptions can occur as a result of pathological nodes, which are considered to exist in all game trees and can cause a deeper game-tree search, resulting in worse play. To reduce the impact of pathological nodes on the search quality, we propose an iterative optimal minimax (IOM) algorithm by optimizing the backup rule of the classic minimax algorithm. The main idea is that calculating the state values of the intermediate nodes involves not only the static evaluation function involved but also a search into the future, where the latter is given a higher weight. We experimentally demonstrated that the proposed IOM algorithm improved game-playing performance compared to the existing algorithms.

Parallel Computing of PVS Node based GTS Algorithm on GPU

The developments in the field of computer architecture, it allows humans to play games like Chess, Tic-tac-toe, Go, etc. with computer machines using AI technology. In AI, Game tree search (GTS) is an important approach and is directed toward finding the finest choice of move for computer games. Using the traditional GTS algorithm the computer could not win a human. So there is a need for enhancing algorithms using dynamic parallelism of GPU. The block of thread set is organized on GPU is either one or two or three dimensional for parallel computations. On this, each thread designated by an inimitable mixture of indices. In this paper, parallel computing of node-based Principal Variation Search (PVS) GTS algorithm presented, which runs on GPU using libraries of CUDA. this experiment tested on chess games with different depths and results are compared with the threads on CPU and GPU. The results proved that GPU improves the performance of speedup up to 80 percent on the checkers game. Parallel computing greatly increasing the efficient use of CPU and improves the performances of the PVS-GTS algorithm on GPU to search deeper layers and find the optimal moves for the current players for two-player computer games.

Development of an Incremental Pattern Extraction Based Gomoku Agent

The subject of this paper is an unusual approach to artificial game playing. Our main goal is to replace exhaustive game tree search with incremental pattern extraction and recognition, thus greatly reducing computation time. This is achieved using search with a depth of 3, together with pattern matching and pattern-based heuristic functions, where patterns are learned through play. We examine the efficiency and efficacy of this method regarding the game Gomoku, also known as Five-in-a-row. To evaluate our agent, we implement two basic reference agents and also incorporate a strong open-source AI called "Carbon" into our environment.